Theft and Loss of Electricity in an Indian State 1 Miriam Golden University of California, Los Angeles Princeton University magolden@princeton.edu Brian Min University of Michigan brianmin@umich.edu April 9, 2012 Version 3.0 Comments welcome. Graphics require printing in color. 1 An earlier version of this paper was presented at the 2011 Annual Meetings of the American Political Science Association, September 2 5, Seattle and at the 2 nd IGC-ISI India Development Policy Conference, December 19 20, 2011, ISI Delhi Center. For research assistance, we thank Julia YuJung Lee. Funding was provided by the International Growth Centre and the Center for International Business Education and Research at the University of California at Los Angeles. Golden acknowledges the hospitality of the Center for the Study of Democratic Politics, Princeton University. The authors are solely responsible for the views presented here.
Abstract Utilizing data from the power corporation of Uttar Pradesh, India s most populous state, we study the politics of electricity theft over a ten year period (2000 09). Our results show that electricity theft is substantial in magnitude. The extent of theft varies with the electoral cycle of the state. In years when elections to the State Assembly are held, electricity theft is significantly greater than in other years. Theft is increasing with the intensity of tubewells, suggesting that it is linked to unmetered electricity use by farmers. Incumbent legislative members of the state assembly are more likely to be reelected as power theft in their locality increases. Our interpretation of these various results is that power theft exhibits characteristics consistent with the political capture of public service delivery by local elites. Our results fail to substantiate that theft is linked either to political criminality or is the product of weak institutions.
1 Introduction In many poor countries economic growth is hampered by inadequate and irregular supplies of electricity. Indian firms ranked electricity problems as the most important issue facing their businesses in the 2006 World Bank Enterprise Survey. The scarcity and unpredictable supply of electricity are in part results of widespread theft, as well as lack of adequate generating capacity. Given its high value, the relative ease with which it is diverted, and the difficulty of identifying individual offenders, theft of electrical power is easily accomplished as well as useful to enterprises and individuals. As a result, it is widespread across much of the developing world. Power theft leads to lost government revenues, reducing the ability of the public sector to pay for the maintenance of existing facilities or to invest in new power generation; it places unexpected strains on already taxed and often inadequate infrastructure, increasing the risk and frequency of power shortages; and it reduces the availability of electricity to paying businesses and consumers. Where power is scarce, firms, agricultural enterprises, and middle-class consumers may offer bribes to government officials to divert electricity illegally, or they may opt out of public sector energy delivery and install their own power generators, creating a dual-track economy. The former potentially establishes persistent collusive and illicit ties between consumers and government officials, whereas the latter reduces the stream of revenues to government. If it is extensive, collusion between government and consumers provides a political incentive to keep electricity supplies inadequate so that government officials may continue to collect bribes. Estimating the extent of electricity theft, the nature of any illicit ties between politicians, power sector bureaucrats, and users, and the political, sectoral and geographic characteristics of users involved in theft is thus one step towards identifying strategies that may ultimately reduce it to manageable levels. We report results of an analysis of electricity theft in Uttar Pradesh (UP), India s most 1
populous state. Using local data on power generation and payment receipts over a ten year period from the Uttar Pradesh Power Corporation Ltd. (UPPCL), the state s electricity provider, we analyze the politics of where and when power theft occurs, the types of users involved, and whether power theft appears linked to other criminal activities. Our analysis is guided by considerations of political economy. We want to know whether power theft is affected by elections, political parties, and the criminal status of state legislators. The reasoning behind our analysis is that the political system controls the institutions that ultimately prevent (or permit) the occurrence of large-scale power theft. In some settings, institutions appear to be relatively effective in preventing such abuses. For instance, widespread power theft is neither a known and noticeable problem in North America or western Europe, nor in some developing countries. In these environments, power use is metered down to the individual household, it is difficult to tap into an electricity line illegally, and bills are regularly issued for power used. Bills that remain unpaid result in a suspension of service. For the interactions of the power corporation and consumers to be vastly different, as is the case in India, things must be different at multiple points in the process. We seek to identify the specific aspects of the system of energy transmission and bill collection that are vulnerable to malfeasance or leakage. The most visible indication of energy theft occurs when users illegally tap into the public supply. Throughout the less developed world, users without access to electricity tap illegally into existing lines, as illustrated in the photograph displayed in Figure 1. Unsanctioned connections to the grid are probably the most frequent way that electricity is stolen. These illegal connections are common and easily detached when monitors or bill collectors arrive, although in some cases they are allowed to remain for indefinite periods. But although they are highly visible and very frequent, illegal hookups are unlikely to be the largest source of energy loss. This instead stems from the two other main ways that energy is sent out but not paid for: meter fraud and unmetered use. One way that 2
Figure 1: Illegally Tapping into the Power Line Source: <http://news.bbc.co.uk/2/hi/business/4802248.stm>. Copyright BBC. meter fraud occurs is when the public utility meter reader is bribed to report an inaccurate number, thereby effectively providing unpaid power to a consumer. This type of fraud is likely to be especially common in settings that are generally characterized by high levels of bureaucratic corruption, although as far as we are aware, there are no accurate estimates of its frequency. Meter tampering is a second type of meter fraud that allows users more power than is paid for. Standard electro-mechanical meters use a slowly spinning disk to record the amount of power that is being drawn. The rotation of the disks can be slowed using magnets or by impeding the disk mechanism with foreign objects depositing spiders and spider eggs to encourage web buildup is one known technique. Disk rotation can be stopped completely by inserting small rocks, gum, or other obstructions. However, 3
complete obstruction of a meter is more likely to attract attention. The third way that energy is lost is through excess unmetered use of electricity. Power is unmetered in various settings. In urban areas, individual apartments may be unmetered, with only a single meter serving a multifamily dwelling. Unmetered use is even more common in rural settings, where it may be difficult and expensive to install individual meters and even more problematic to ensure that they are regularly and accurately read. In place of meters, farmers pay flat rates calibrated to the size of their equipment. As a result, as Varshney (1998, p. 171) contends, agricultural consumers... account for approximately 25 percent of total electricity consumption... and are responsible for the bulk of the power sector s financial losses. More recently, others have deepened the criticism, arguing that thanks to perverse subsidies under its flat system of electricity pricing, India s booming groundwater irrigation economy has wrecked its energy economy (Shah, Giordano & Wang 2004, p. 3452). India deliberately abandoned metering the power supply for agricultural irrigation in the 1970s, as part of the Green Revolution strategy of switching to new high-yield crops, which required regular water supplies. The provision of subsidized power to farmers was considered a critical investment for improving the productivity of the agricultural sector. Since the 1970s, Indian agricultural irrigation has involved flat rate connections with tariffs set at the state level; the tariffs vary with the horsepower of the energy-utilizing equipment. In the absence of technical enforcement mechanisms, the temptation to overdraw electrical power beyond the contracted flat rate level can be high. In the agricultural sector, electricity is used to power irrigation pumpsets or tubewells to extract groundwater for crop watering. In many parts of the country that cannot rely on rainfed crops, including Uttar Pradesh, low cost power represents one of the most significant and expensive subsidies to the farming sector. Part of the subsidy is publicly visible, effected through the pricing schemes adopted by government, but another part is 4
invisible, occurring when enough electricity is sent out to irrigation equipment such that farmers are permitted to use more than the maximum amounts they are officially allocated. In South Asia, some 14 million electric tubewells pump water mainly for irrigation purposes without being metered (Shah, Scott, Kishore & Sharma 2004, p. vi). Hence, the basic features of the environment that we study in UP are common to the region. The main results of our analysis are as follows. First, we corroborate that power theft in India in large in magnitude (Transparency International India 2005). We document that in UP, theft is greatest in periods immediately prior to state elections. Extending this line of argument, we document that incumbent members of the state assembly are more likely to be reelected in areas where power theft is more extensive. Power theft, we show, is most intense in the state s most agricultural localities, suggesting it is largely due to unmetered agricultural use. The natural interpretation is that farmers simply exceed their allotted maximums when more energy is supplied. Perhaps as a result, power theft is not related to specific markers of political criminality. We find little in our data suggestive of persistent collusive illegal networks linking politicians and users. While we know that individual meter readers accept bribes to underreport the amount of electricity used, the aggregate effects of this appear relatively small. Instead, our findings are consistent with the view that relatively well-off farmers those who own electric tubewells comprise a powerful interest group to whose interests democratically elected state legislators are particularly sensitive. As those farmers with the wealth to own tubewells comprise a rural elite, our results are consistent with a literature that argues that public service provision is liable to political capture by local elites (Bardhan & Mookherjee 2000). Our paper proceeds in seven parts. First, we briefly review related literature. Second, we present four sets of hypotheses that we study. Third, we describe some basic characteristics of electricity use in Uttar Pradesh and, fourth, we describe our dataset. A fifth section provides descriptive statistics and a sixth, the results of statistical estimations of our four 5
sets of hypotheses. A final section offers concluding thoughts and hypotheses for future research. 2 Related Literature Our paper is related to studies of the political business cycle in subnational units (examples include Baleiras & Costa (2004), Drazen & Eslava (2005), Mouriuen (2007)), which grew out of studies of the political business cycle at the national level (Nordhaus 1975, Tufte 1980). Various papers show that municipal level elected officials manipulate aspects of the local political economy prior to elections in order to improve their chances of reelection. Of particular relevance is Khemani (2004), which documents state-level electorally sensitive targeting of advantage to special interests in India. Also important for our purposes is Shi & Svensson (2006), which finds that the political business cycle is larger in less developed than in developed countries, suggesting that elected officials are under greater pressure to manipulate the economy prior to elections in poorer countries. This may take unusual forms that would not be encountered in developed economies. For instance, Burgess, Hansen, Olken, Potapov & Sieber (2011) identity political logging cycles in Indonesia, where illegal logging increases substantially in the years prior to local elections. A large related literature on political corruption is also relevant (Rose-Ackerman 1999, Johnston 2006, Treisman 2007), especially studies that document that corruption rises or declines according to the reelection incentives of local politicians (Ferraz 2006). The only study of which we are aware that specifically studies energy theft as a problem of corruption is Smith (2004). This cross-national study of transmission and distribution (T&D) losses in energy transmission finds that the extent of such losses is highly correlated with corruption in general, as well as weaknesses in accountability and institutional performance. 6
Our study draws on a large literature on the politics of public goods provision, as well as a small literature that studies the political economy of electricity provision in particular. The former is exceedingly vast; for a relevant review, see Golden & Min (Forthcoming 2013). The main result of the distributive politics literature is to underscore that public officials use electoral criteria in the allocation of public and government goods and services rather than utilizing strictly welfare maximizing criteria. There is considerable national and local variation in how this occurs, however, in part because features of electoral competition differ. As regards electricity provision, Brown & Mobarak (2009) show that in poorer countries, democratic political institutions shift electricity provision from the industrial sector to households, whereas authoritarian institutions favor industry. Min (2010) documents partisan effects in electricity provision in Uttar Pradesh. Other studies, including Bernard, Gordon & Tremblay (1997), show that electricity prices may be politically manipulated for electoral ends, in line with the general distributive politics theme. In a paper especially related to this one, Badiani & Jessoe (2011) show that the well-known price subsidies to Indian agriculture for electricity are partially due to political capture; subsidies increase significantly in the year prior to an election. Finally, our study is informed by a theoretical literature on policy distortions due to capture by special interests (Grossman & Helpman 2002, Bardhan & Mookherjee 2000). 3 Hypotheses Explored in the Study We seek to understand variations in line loss across sectors, geographic units, and years in Uttar Pradesh. Drawing on existing literature for our hypotheses, we use a variety of statistical techniques to study the following four questions: 1. Does line loss increase in electoral periods? 7
2. Is line loss greater for important socio-economic interests, especially agriculture? 3. Is line loss electorally beneficial to state assembly members? 4. Is line loss greater in geographic areas whose elected representatives are under selfreported criminal indictment? We provide details about the specific estimation techniques used for each of these later in the paper. After presenting our results, we specify some mechanisms that could explain our findings. 4 Electricity in Uttar Pradesh Uttar Pradesh is India s largest state, with a population of 190 million people in an area about half the size of California. If it were a country, it would have the fifth largest population in the world. According to World Bank estimates, it is home to 8 percent of the world s poor. As the map presented in Figure 2 shows, UP sits in the center of northern India. All electricity transmissions and distribution in the state is controlled by the Uttar Pradesh Power Corporation. The UPPCL was formed in 2000 as a result of power sector reforms and the unbundling of the state electricity boards across India. UPPCL remains a state-owned entity. Its workers are state employees and its key leadership positions are filled by political appointees. The managing director of UPPCL is drawn from the Indian Administrative Service, the country s elite corps of public sector managers. Compared to a baseline estimated demand of between 7.5 and 9 gigawatts (GW), UP- PCL is capable of providing up to about 6 GW of power at any point in time. For comparison, this is roughly the level of electricity consumption of the state of Connecticut, whose population is about 2 percent that of the population of Uttar Pradesh. Electrical 8
Figure 2: The State of Uttar Pradesh in India Uttar Pradesh power in UP is distributed through an intricate network of generating plants, substations, transformers, and thousands of miles of power lines. To manage the surplus demand and protect the fragile power grid, electrical power has to be rationed and massive blackouts sweep across the state every day of the year. At any given time, one-fifth of users are typically without power. Standard guidelines exist for the scheduling of blackouts, although these guidelines are frequently not adhered to. Urban areas are supposed to get 20 hours of power a day and villages, 12. The UPPCL in fact exercises considerable discretion in the transmission of electricity to localities, and is under constant pressure from consumers 9
(and elected officials) to provide power when supply is inevitably inadequate. Villagers regularly complain that electricity supply occurs unpredictably and that the number of hours of power they receive is much less than those promised. For the UPPCL, consumers are distinguished by sector. In 2008, there were 10 million consumers registered with the UPPCL. Of these, 81 percent were domestic, 10 percent commercial, 1.5 percent industrial, and 8 percent agricultural. These groups of consumers vary greatly in their intensity of use. In terms of connected load, 55 percent went to domestic users, 9 percent to commercial users, 16 percent to industry, and 17 percent to agriculture. Thus, the average agricultural user was connected to three times the load of a typical domestic consumer. The composition of billing for electricity use varies yet again because of differences in tariffs across sectors. In 2008, to again use the year for illustrative purposes, domestic users accounted for 29 percent of the total amount billed, 12 percent went to commercial customers, 45 percent of bills went to industry, and a mere 5 percent of billing went to agricultural users. As these figures show, agricultural users enjoy a large subsidy, paying for 5 percent of total electricity while accounting for 17 percent of the total electrical load. Industrial users, while connected to a similar share of total load as agriculture, pay 45 percent of the total amount of electricity billed. These figures suggest what we further document below: agricultural interests are favored in electricity allocation. At least in part, tariff regimes are subject to political manipulation and can be adjusted in order to secure the electoral support of different constituencies. In mid-2006, the ruling Samajwadi Party (SP) government announced a lower flat rage tariff structure for power looms used by weavers in the state. The new rates were 65 Rupees ($1.44) per horsepower in urban areas and 37.5 Rupees ($0.83) per horsepower in rural areas, providing weavers access to power at the same low tariffs as the powerful farming sector. 1 The decision 1 Power at flat rate to U.P. weavers, The Hindu, 21 June 2006. http://www.hindu.com/2006/06/21/ 10
was notable for how finely targeted the beneficiaries were: there are only about 300,000 power looms in the state, concentrated in the districts of Mau, Varanasi, Ambedkar Nagar, Meerut and Jhansi. The timing of the decision also appeared to be politically motivated, announced just months before the 2007 state elections. Finally, many weavers are of the Kori caste, one of the Scheduled Castes who form a critical element in the core support base of the opposition BSP party. It would not be implausible to hypothesize that the SP s subsidy was an attempt to wrest from the BSP the electoral support of voters who owned power looms. Likewise, prior to the 2012 state assembly elections, the SP, which received a large plurality of votes, announced that if elected it would implement a policy of free electricity to the state s weavers and farmers. 2 5 The Data We collected administrative data on electricity use from the Uttar Pradesh Power Corporation Ltd. from 2000 to 2009. 3 The availability of data is the main reason that we selected UP for analysis, although its large size makes it a prominent and important case. Moreover, it is worth noting that, according to Transparency International India s ranking of corruption across 20 major Indian states, UP falls right in the middle (Transparency International India 2005, table 1.5, p. 10), making it broadly representative of the country as a whole. In India, public electricity providers, which are state-specific, are widely viewed by the public as corrupt (Transparency International India 2005, p. 49). Our primary outcome variable is line losses, measured as the share of electrical power that is distributed from the power station but not billed for. In many contexts, line loss is stories/2006062107500400.htm 2 SP to waive farmers loans if voted to power in UP, Business Standard, 16 Jan 2012. http://http://www.business-standard.com/india/news/sp-to-waive-farmers% 5C-loans-if-voted-to-power-in-up/155395/on 3 The data are recorded monthly, though we focus on annual fiscal year totals in this paper. 11
known as transmission and distribution (T&D) losses. Some line losses unavoidably result from technical factors. Over long distances, power inevitably degrades due to physical factors inherent to the transmission process. Such technical losses range from 1 2 percent in efficient systems to as high as 9 12 percent of total power output in less efficient systems (according to Smith (2004, p. 2070)). Line losses in India are much larger than this, on the order of 30 percent. As we noted above, the larger share results from meter tampering, bypassing of meters via illegal connections, and unauthorized excess usage by flat rate customers. We call the share of power that is used but unpaid for, theft, although part of this comprises genuine T&D losses. Even if we grant that as much as 12 percent of line loss may stem from technical features of India s inefficient power system, theft itself comprises a total amount that is fifty percent greater than this. Line losses are not the only losses experienced by the UPPCL. Even when bills are sent to customers, many go unpaid, aggravating the power company s revenue shortfalls. Bills go unpaid for numerous reasons, only some of which might plausibly be related to corruption by corporation officials or to deliberate consumer malfeasance. Bureaucratic inefficiencies might prevent the collection of bills. Even for those willing to pay, making payments in India can be difficult. Because it has not been possible until extremely recently to pay electricity bills electronically, consumers must pay in person at a UPPCL office. In remote rural areas, customers must often travel long distances to pay their bills. Because we believe that much of the non-payment of bills is due to factors such as these (but we have no way to estimate the proportion), we do not use non-payment as a proxy for electricity theft, even though the result of non-payment is effectively such. The power company collects and reports data at the level of the geographic service division, which are units specific to the UPPCL. The state of Uttar Pradesh was divided into 193 divisions at the end of 2009. When the number of customers within a division gets too large, the division is split. As a result, the number of divisions at the beginning of 12
our time frame is smaller than in 2009. In our analysis, we aggregate divisions that were split back to their 2000 boundaries in order to create a uniform series. This results in 154 observations for analysis. Additional administrative data records the number of consumers, the total connected load, and total billing, broken down by sector (residential, commercial, industrial, and agricultural, among others) and by division. The true usage by different types of users or individual consumers is not known, only the total supply delivered from each power substation and the total amount that is billed for. The gap between power that is delivered and power that is billed represents line losses. This data enables us to describe the composition of consumers within each division, thus identifying areas whose intensity of energy use is more agricultural or more industrial, for example. However, line losses can only be estimated at the division level and cannot be further disaggregated by sector; that is, we do not have the information to report the precise proportion of line loss due to agriculture, industry, households, or commerce. Because we are interested in the possible political correlates of power theft, we collect data on a number of potentially relevant political factors. The first are state assembly elections. Electricity provision is a state-level responsibility in India s federal structure, power company officials are state employees, and key appointments to the power company leadership are made by elected state leaders. Village leaders have little ability to influence the provision of electricity to their localities. Thus state assembly elections are the most salient level for political analysis, more than federal parliamentary elections or village council elections. Uttar Pradesh has 403 single-member state assembly constituencies and elections to the Vidhan Sabha, its lower house, were held in 2002 and 2007. The 1990s were characterized by intense electoral competition and fragile coalition governments formed between new parties that had helped crack and supplant the Congress Party from its decades-long grip on power in both the national capital and UP s state cap- 13
ital, Lucknow. Prior to the 2002 election, the Chief Minister s office (equivalent to a state governor in the United States) was held by the Bharatiya Janata Party (BJP), a conservative Hindu nationalist party with strong support from upper caste and middle-class urban voters. The BJP was in the process of strengthening its claim as the most powerful party in post-congress India. However, the 2002 UP state elections dealt a severe blow to the BJP s upward trajectory, as it won fewer seats than both the Bahujan Samaj Party (BSP) and the Samajwadi Party. The BSP s core support came from Scheduled Castes comprised of groups who historically occupied the very lowest rungs of India s social hierarchy while the SP enjoyed the support of many Other Backward Class (OBC) and Muslim voters. In the 2007 elections, the BSP won an outright majority of seats in the state house, the first time in two decades that coalition rule was not required. The success of a party that championed the interests of UP s poorest and most marginalized citizens was both a stunning and unexpected achievement. Our data track this period of deep political and social transformation in Uttar Pradesh. A second political factor that we incorporate into our work is the self-reported criminal status of candidates to the UP State Assembly in 2007. In 2003, the Indian courts issued a ruling requiring that all federal and state level legislative candidates provide sworn affidavits in which they reported, among other things, whether they were currently under criminal indictment or had been convicted of criminal malfeasance. The timing of the court ruling is such that this information is unavailable for candidates to the 2002 State Assembly. However, the information is available for the 2007 elections. We utilize it for the 403 assembly constituencies, which saw just over 6,000 candidates run, or an average of 15 per constituency. Of these, approximately 11 percent of candidates were either convicted criminals or had criminal charges pending against them. However, of the 403 legislators elected in 2007, fully 25 percent were either under criminal indictment when elected or had previously been convicted of criminal malfeasance. Although we do not have infor- 14
mation on the nature of the charges, it is reasonable to investigate whether power theft is greater where legislators with criminal records or facing indictment hold the seat. There is no way to directly map the 403 assembly constituencies to the 170 geographic service divisions, since boundaries of the UPPCL service divisions are not published. Each assembly constituency and UPPCL service division can, however, be precisely located within a single administrative district, which is a unit roughly comparable to a U.S. county. We can thus aggregate data from both other levels to the administrative district level, of which there are 70 in Uttar Pradesh. In addition, census data (from 2001) are available at the level of the administrative districts. We therefore are able to merge into our dataset a range of relevant control variables at the level of administrative districts. Given the mismatch in the geographic levels between our power theft variables and our political variables, there is no single optimal way to merge the data together for analysis. One option is to aggregate all the data into larger units, computing averages and totals at the level of the 70 administrative districts. However, we lose a lot of information doing this. We can also create a separate dataset at the assembly constituency level (but with imputed electricity data drawn from the UPPCL service zone) and another at the UPPCL service division level (but with imputed electoral data from the assembly constituency). These alternative scenarios lead us to construct three datasets, one at the administrative district level (n = 70), a second at the UPPCL service division level (n = 154), and a third at the assembly constituency level (n = 403). We utilize each of these for different parts of the analysis. At every step, we utilize whichever dataset has the largest number of relevant observations, given the specific question. The UPPCL division level dataset allows us to describe characteristics of power use and theft at the most detailed level, while estimating political effects from electoral constituency data aggregated to the larger district in which the division is located. We use this dataset to examine where power theft is greatest and the characteristics of politicians 15
elected in the districts in which the division is located. The assembly constituency level dataset is most appropriate for exploring determinants of election outcomes as well as the criminal status of assembly candidates. With these variables, we can examine whether politicians are more likely to win when their constituency is in a district with higher rates of power theft and whether tainted candidates appear more often in constituencies with more power theft. Finally, we use the administrative district dataset, which contains the most aggregated data, to evaluate the robustness of our findings. 6 Descriptive Analysis Nearly a third of all electrical power in Uttar Pradesh is unaccounted for. In other words, adding up all the meter readings from all consumers in the state only results in bills that amount to two-thirds of the power sent out by UP s power stations. The remaining power cannot be tracked and is assumed lost to ordinary T&D losses as well as to theft, meter tampering, and excess usage by flat rate customers. The proportion of power that is lost in UP is approximately the same as the national average (Narendranath, Shankari & Rajendra Reddy 2005, table 3, p. 5566). 6.1 Geographic Variations There is wide variation in electrical line losses across Uttar Pradesh. In 2005, for example, a stunning 66 percent of all power in the Mainpuri district was not billed for. Meanwhile, in that same year, line losses were lowest (just under 13 percent) in the Sonbhadra district. 4 Line loss is, as we observe from the data depicted in the upper panel of Figure 3, greatest in the western part of the state and generally less farther east. This difference coincides 4 Sonbhadra is sparsely populated and home to several of India s largest coal-based thermal power plants. 16
with the differential distribution of tubewells in the state, whose irrigation coverage is 27 percent greater in western than in eastern UP (authors calculations using 1998 99 figures reported in Pant (2004, p. 3464, Table 1)). For comparison, the lower panel of the figure shows a satellite-based image of nighttime light output, which depicts variations in the availability of power and intensity of use (Min 2010). 5 The image is a composite of all satellite imagery captured of Uttar Pradesh between 20.00 hrs and 21.30 hrs local time across the calendar year. Further processing excludes images shrouded by cloud cover and other digital noise. The composite image shows no obvious correlation between overall electricity use and the rate of line losses. This supports the view that most line loss is due to factors other than merely technical features of the transmission and distribution of electricity. Table 1 lists the districts with the highest average line losses between 2000 and 2009. On average, half of all power supplied in the Hathras district (now known as Mahamaya Nagar) could not be accounted for, higher than any other district in the state. Among the other leading districts, Etawah is the home of Mulayam Singh Yadav, leader of the Samajwadi Party and Chief Minister of the state from 2003 to 2007. Mainpuri is home to his brother and a stronghold of the Singh Yadav family. The districts with the lowest line losses on average during our study period are listed in Table 2. At the top of the list is Gautam Buddha Nagar, home to the bustling outsourcing hub of Noida, just east of New Delhi. The efficiency of collections in this district may reflect a greater willingness to bill commercial customers, including many foreign-owned entities. 5 Analysis in Min (2010) shows that nighttime light output and electricity consumption at the district-level are very highly correlated in Uttar Pradesh. 17
Figure 3: Linelosses and Nighttime Lights Across Uttar Pradesh Note: Line losses in districts in fiscal year 2005. Average evening hour nighttime light output from 2003. Sources: UP Power Corp, US Air Force Weather Agency, and NOAA-NGDC. 18
Table 1: Highest Line Losses by District, 2000 09 Average District Line losses (%) Energy Supplied (MU) Energy Billed (MU) Hathras 49.9 472.5 192.7 Mainpuri 49.9 241.7 118.5 Jhansi 45.8 662.2 364.8 Jalaun 45.7 419.2 231.9 Etawah 45.4 321.8 173.5 Bulandshahr 43.8 933.0 526.5 Saharanpur 42.8 1233.9 709.4 Firozabad 42.5 675.5 395.7 Rampur 42.3 370.7 216.6 Moradabad 40.5 964.1 573.2 Table 2: Lowest Line Losses by District, 2000 09 Average District Line losses (%) Energy Supplied (MU) Energy Billed (MU) Gautam Buddha Nagar 13.6 1370.0 1197.0 Sonbhadra 16.4 259.7 218.1 Lakhimpur Kheri 19.5 218.2 174.8 Basti 19.8 196.7 157.4 Kushinagar 20.0 142.2 113.1 Maharajganj 20.3 120.7 95.8 Deoria 20.7 211.2 166.5 Hardoi 21.9 252.4 195.6 Sitapur 22.6 211.8 163.2 Hamirpur 22.8 275.9 213.3 6.2 Variations in Line Loss Over Time In Uttar Pradesh, line losses have been decreasing over time, as documented in Table 3. From a rate of 37 percent in 2000, total line losses has declined steadily to just under 25 percent in 2009. Two modest peaks in the downward trend occur in 2002 and 2007, which correspond to election years. 6 The higher averages in election years seem to result from higher losses in UPPCL di- 6 The UPPCL fiscal year runs from April to March. Elections were held in May 2002 and February 2007. 19
Table 3: Average Line Loss by Year Across UPPCL Divisions, 2000 09 Average Fiscal year Obs %Lineloss Std. Dev. Min Max ------------+--------------------------------------------------- 2000 154 37.02 12.42 3.97 65.32 2001 153 35.26 12.15 2.26 63.38 2002 146 36.85 9.98 13.33 62.51 2003 149 28.51 11.39 5.74 59.13 2004 158 35.33 11.67 14.70 69.00 2005 169 31.14 11.42 10.53 65.77 2006 170 31.25 10.56 9.60 64.08 2007 179 31.90 10.04 8.42 65.79 2008 190 29.89 9.46 8.19 63.72 2009 193 24.83 7.84 5.45 50.87 visions all across the state and do not appear geographically concentrated. In Figure 4, we draw kernel density plots showing the distribution of line losses across all geographic observations in each year. The election year lines are shifted towards the right, indicating broad-based increases in losses. 7 Statistical Analyses Thus far, we have identified two patterns in our data of potential theoretical importance. First, line losses are geographically concentrated in western UP, where more tubewells drawing irrigation water are located. Second, although losses have fallen over the period examined, they are greater in years of state elections than other years. We now study these patterns using more systematic methods. As already indicated, we investigate four questions: (1) whether power theft is affected by the occurrence of an election to the State Assembly; that is, whether we observe a political business cycle to line loss; (2) whether theft occurs across all groups in society or whether some appear to engage in more power theft; (3) whether theft pays politically; that is, whether political 20
Figure 4: Kernel Density Plots of Linelosses by Year 21
incumbents benefit electorally from power theft; and (4) whether the geographic areas with more power theft exhibit other symptoms of political criminality. 7.1 Electoral Cycles The annual data just reviewed suggests that electoral effects in line loss may be present. Confirming this, we find strong statistical effects for an electoral cycle in line losses. The t-test reported in Table 4 shows that line losses are nearly 3 percentage points higher in election years than other years and the difference is highly statistically significant. Table 4: UP Line Losses in Election and Non-Election Years, 2000 09 Division-Year Mean Std. Dev. Observations Election Year 325 34.1 10.3 Non-Election Year 1,336 31.4 11.5 Difference +2.7 p 0.001 To investigate how robust these results are, we conduct regression analysis. As we document in the results reported in Table 5, the election year effect is slightly larger in fixed effects regressions that study line loss with election years as regressors and that include indicator variables for each division to control for time-invariant division characteristics. These results provide circumstantial evidence that the provision of electricity is subject to political manipulation. Constituents appear to benefit from reduced efforts by the state to monitor electricity use in periods prior to elections. 7.2 Who Steals: Sectoral Analysis The UPPCL geographic divisions vary in the share of power that goes to different sectors, including the domestic, industrial, commercial, and agricultural sectors. Given the rela- 22
Table 5: Fixed Effects OLS: Testing for Electoral Cycles in UP Line Loss Variable Coefficient (Std. Err.) Election Year 2.847 (0.391) Division Fixed Effects Yes Intercept 30.972 (1.554) N 1661 R 2 0.675 Significance levels : : 10% : 5% : 1% tive economic and political importance of these varying constituencies, we examine how line losses varies with the sectoral mix of customers within each geographic zone. Our data permits us to describe whether line losses are higher in places with more domestic customers, more industry, or more agriculture. (Recall that the data do not permit us to identify sectoral line losses per se.) Since we are interested in whether politicians sanction energy theft for electoral gain, identifying who is allowed to steal is relevant. Figure 5 plots line losses in 2007 compared to the proportion of energy load going to different sectors. Each point represents a geographic service division. The most notable pattern is that line losses are increasing in private tubewells, but flat or decreasing in the other three sectors that we depict (domestic users, industry, and commerce). This means that as the proportion of electricity to private tubewells increases, the proportion of energy lost out of the total sent rises. Hence, where agriculture is a more concentrated interest, there is more power loss. In addition to the fact that private tubewells are by and large used for agricultural irrigation, we note that they are markers of relatively wealthy agricultural interests, since only relatively well off farmers are able to afford the financial burden of installing and maintaining a tubewell. Although electric tubewells are by far the most efficient way to draw water in UP (much more efficient than diesel-generated tubewells, for instance), they are very expensive by local standards. As a result, we estimate that only 23
1.5 percent of farmers in UP own electric tubewells. 7 Figure 5: UP Line Losses by Sector Why are line losses higher in more agricultural areas? Agriculture is the largest economic activity in Uttar Pradesh, accounting for nearly half of gross state product in 1991 and employing three-quarters of the labor force. Farmers are among the most important of electoral constituencies in the state. We interpret this result as indicating the tacit willingness of the state government to ignore electricity theft by relatively wealthy farmers, especially in election years. Electricity theft in this instance refers to the transmission 7 The UPPCL data reports 729, 000 private electric tubewells in the state as of April 2008 (there are another 2.7 million estimated tubewells using diesel fuel in UP). Indian census data from 2001 report 49.3 million cultivators in the state. Thus, there is approximately one privately-owned electric tubewell for every 68, 000 cultivators. 24
of more electricity to tubewells, which run on a separate part of the grid, than would be appropriate given their allotment. Our results are consistent with an Indian cross-state analysis by (Joseph 2010), which shows that private sector exit from public utility electricity generation between 1994 and 2005 is significantly greater with more agricultural pumpset consumption and higher T&D losses, among other variables. He interprets these findings as showing that industrial consumers set up their own generating plants more often where the quality of electricity is poorer due to greater unmetered agricultural use. As unmetered agricultural users are offered more electricity, the supply to other sectors must necessarily be interrupted more frequently, due to the overall scarcity of power. In our conclusions, we offer hypotheses that specify more precise mechanisms linking the interests of the state s wealthiest farmers to state political interests. We also compare the slopes of the best-fit line for private tubewells in 2007 with the year before and after; our results are presented in Figure 6. The slope is higher for the election year, which is again suggestive that political intervention to enable or disregard theft is occurring. (A similar analysis, not reported here, for 2002 found no election year effect, however.) 7.3 Incumbency Effects A third pattern that is consistently strong in the data we analyze is the relationship between reelection and line losses. Previous research has documented a significant incumbency disadvantage for Indian state legislators (Uppal 2009). The February 2007 elections appear consistent with this expectation: across UP s 403 state assembly constituencies, only 146 of the winners were incumbents who had served in the same seat in the prior legislature. Notably, the rate of line losses was for these incumbents was 33.0 percent compared to 30.8 percent for all other legislative constituencies. At first glance, it does appear that incumbents may benefit electorally from higher line losses within their constituencies. 25
Figure 6: Linelosses in Agriculture, by Year We explore whether this difference in line losses could help explain variations in reelection rates. Since incumbency re-election is a function of several factors, we run a logistic regression on whether an incumbent was re-elected in the February 2007 elections. The main theoretically relevant independent variable is the measure of line loss in the assembly constituency in fiscal year 2007 (April 2006 March 2007), most months of which occurred prior to the election. We also include as control variables whether the constituency seat is reserved for a member of a Scheduled Caste, 8 three measures of economic welfare from the UP Human Development Index (average income, education, and health), the size of the electorate, the turnout rate, and controls for which party controlled the seat in the prior period. In the logistic regression results reported in Table 6, the results document a positive 8 There are no assembly seats reserved for members of Scheduled Tribes in UP. 26
Table 6: Logit Regression Predicting Incumbency Reelection February 2007 UP State Assembly Elections Variable Coefficient (Std. Err.) Line losses, FY 2007 0.028 (0.016) Criminal charges 0.056 (0.286) Reserved SC seat -0.137 (0.311) HDI 6.138 (2.933) Number of voters -6.446 (3.176) Voter turnout -4.535 (2.102) Win margin, last election 8.172 (1.839) Seat previously held by SP 1.882 (0.546) Seat previously held by BSP 3.564 (0.573) Seat previously held by BJP 1.433 (0.573) Seat previously held by INC 1.180 (0.780) Intercept -3.713 (2.175) N 392 Log-likelihood -202.285 χ 2 (11) 104.011 Significance levels : : 10% : 5% : 1% and marginally significant effect of line loss on the probability of being reelected. 9 One possible interpretation of our results is that when incumbent MLA s allow high rates of power theft, they are more likely to be rewarded by voters and be reelected. 7.4 Criminal Environments With a quarter of the members of the UP state assembly reporting that they are either under criminal indictment or have previously been convinced on criminal charges, there is evidence of criminal intrusion into state politics. In this section, we study whether line losses are significantly greater in assembly constituencies with candidates or elected representatives who report criminal records compared with constituencies with no candidates or no representative reporting criminal malfeasance. 9 Again, we observe no similar electoral effect in 2002. 27
All candidates in the 2007 election were required to file affidavits stating whether they faced any pending criminal charges. We analyzed data for 6,055 candidates across the state. Of these, 645, or nearly 11 percent of candidates reported criminal charges. Criminal candidates competed in 288 of the 403 constituencies, with a criminal candidate winning 103 of these seats. Overall, we find no confirmatory evidence that line losses differ substantially in areas with criminal candidates. Figure 7 compares criminality against line losses at the district level in 2007. The x-axis shows the proportion of candidates facing criminal charges in each district while the y-axis shows line losses in the district. No apparent relationship emerges between line losses and environments that attract criminal candidates. Figure 7: Linelosses and Criminality of State Assembly Candidates, Districts in 2007 We examine the relationship more closely with a regression that adds several districtlevel controls including level of development (HDI index), population, proportion SC, and proportion urban. The results, shown in Table 7, confirm the lack of a relationship between 28
Table 7: OLS Regression on District-level Line Losses, 2007 Variable Coefficient (Std. Err.) Prop. Criminal Candidates 3.581 (19.408) Level of Development (HDI) 45.681 (28.661) Population -2.618 (1.239) Prop. Scheduled Caste -37.712 (20.266) Prop. Urban -2.337 (12.517) Intercept 21.356 (15.612) N 68 R 2 0.175 F (5,62) 2.634 Significance levels : : 10% : 5% : 1% criminal candidates and line losses. By contrast, line losses are lower in districts with larger populations and more voters who are members of Scheduled Castes. The results are unchanged if we look only at the share of winning candidates that are criminals in each district (not reported). One interpretation of these results is that self-reported criminal politicians are not linked in systematic ways to line loss. If they were, we might suspect that they were accepting kickbacks from users in order to permit power theft. That instead they are not suggests that power theft, although of clear electoral benefit to incumbent MLA s, is not part of persistent illicit criminal networks linking elected politicians and users. However, as displayed in Figure 8, realized revenue, which is a measure of bills collected relative to bills issued, are substantially higher in clean constituencies. This trend is confirmed in the multivariate regression reported in Table 8. The model predicts that a 10 percent increase in the share of criminal candidates is associated with a 5.4 percent decrease in realized revenues. In other words, places where people do not pay their bills appear to attract state assembly candidates with criminal records. This result is not subject to any clear and unambiguous interpretation. It may indicate an environment of generally high criminality, or both high line loss and high rates of criminal candidates may instead reflect other phenomena, such as a tight connection between the ownership of pri- 29